A Further Study On The Impact Of Parameter Errors On The ENSO Spring Predictability Barrier In The Zebiak-Cane Model

Both initial and parameter errors can cause prediction errors. Previous researchshowed that the initial errors are more important than the parameter errors for thespring predictability barrier (SPB) in the Zebiak-Cane model (ZC model). Althoughthe parameter errors cause much smaller prediction errors than the initial errors, it isstill possible that when the initial and parameter errors coexist in the model, theoptimal combination of these two types of errors can cause a much larger predictionerror than the initial errors alone. In this case, we cannot overlook the importance ofthe parameter errors, because they can stimulate the growth of the initial errors. In thispaper, we consider such situation with the ZC model and choose eight different ElNi o events, including four strong events and four weak events. For each event, wemake predictions for12months with eight different start months. Using theconditional nonlinear optimal perturbation (CNOP) approach, we calculate the CNOPerrors (the optimal combination of the two types of errors), the CNOP-I errors(optimal initial errors with initial errors considered only), the CNOP-P errors (optimalparameter errors with parameter errors considered only), and the CNOP-I+CNOP-Perrors (the simple linear combination of the CNOP-I and CNOP-P errors). The mainaim of this thesis is to investigate whether the CNOP errors can lead to a moresignificant SPB than the CNOP-I errors, so as to demonstrate the importance of theparameter errors. By comparing the growth of different optimal errors, we concludethat:(1) From the perspective of the error seasonal growth tendencies, the CNOPerrors show an obvious season-dependent evolution, except for the prediction startingfrom April. The error growth tendency of the CNOP errors is not significantly greaterthan that of the CNOP-I errors. In addition, we find that the error growth is faster inthe growing phase than in the decaying phase.(2) Although the CNOP errors lead to the largest prediction errors, the prediction error of the CNOP errors is not significantly larger than that of the CNOP-I errors.Also, the strength of the El Ni o events may affect the prediction errors. During thegrowing phase, the prediction errors of the strong events are larger than those of theweak events, while for the decaying phase the weak events have larger predictionerrors.(3) The pattern of the error evolution shows that the CNOP errors and theCNOP-I errors are notably similar to each other in terms of not only their spatialstructures but also their values.The above results show that the CNOP errors cannot cause a much moresignificant SPB than the CNOP-I errors, which further demonstrates that parametererrors are less important than the initial errors for the SPB in the ZC model. Weshould improve the prediction skill by simply focusing on reducing the initial errors inthis model.